Assessing the presence of biasing and non-biasing unblinding in a randomized controlled trial of a home-use tDCS device: An exploratory analysis.

Better-than-chance guessing of treatment assignment raises concerns about unblinding as this opens efficacy estimates up to being influenced by factors other than the true efficacy of the intervention. However, such accuracy does not necessarily indicate that efficacy estimates are biased. To assess whether unblinding was biasing (impacting on efficacy estimates) or non-biasing (not impacting on efficacy estimates) in a sham-controlled trial of a home-use transcranial direct-current stimulation (tDCS) device, in which treatment allocation guesses at endpoint were unbalanced. Interrelations between i) treatment allocation, ii) treatment allocation guesses, iii) adverse event reports, and iv) depressive symptom severity, as measured by Hamilton Depression Rating Scale, after ten weeks of treatment were examined using linear regression (n = 149). Receiving tDCS was positively associated with a) guessing that one had received tDCS (p = .01704), b) number of adverse events (beta: 0.69, p = .0046), and c) improvement after ten weeks (beta: -2.39, p = .0049). Guessing that one had received tDCS was positively associated with greater improvement (beta: -3.16, p = .0005), whereas the number of reported adverse events correlated negatively with improvement (beta: 0.68, p = .0186). Moreover, among participants who received tDCS, those who reported any of nine examined adverse events consistently showed numerically worse outcomes than those who did not report the same adverse events. In this exploratory analysis, adverse events correlated negatively with HDRS-rated improvement. This is not consistent with a pathway where adverse events lead to unblinding and, subsequently, to improvement via expectancy effects and/or biased ratings.